The search for an accurate, at-line real time multi-analyte analyzer has been ongoing for many years. The application space for such an analyzer is both diverse and broad, spanning many fields including but not limited to food and beverage manufacture, medical diagnostics, chemical analysis, energy, and especially bio-processing and bio-technology. While the description of the present invention focuses mainly on biotechnology based applications, the invention described and claimed and the principles discussed in respect thereto are general in applicability and the conclusions are not limited to the biotechnology application arena.
The demand for this type of analyzer instrument in many fields is based on the need to quantitatively analyze multiple analyte species present within a sample in a closed container. Numerous attempts have been made to apply a wide variety of technologies for this purpose. While instruments exist today that invasively take samples which are then analyzed using standard chemical assays in a more or less automated format, the preferred analytical solution would be non-invasive and therefore not require a breach in the wall of the sample container. Some of the most often applied technologies to achieve sample analysis are so-called “label free” optical technologies that probe the system under test. These techniques are minimally invasive although not actually non-invasive and hold the promise of not compromising either the test sample or the system under test.
Linear scatter and absorption based optical analytical techniques have to date proved to be of only limited use, as they either lack specificity or require some tag or label to be introduced into the sample, thereby contaminating it and making it unavailable for further use. Early attempts using optics often employed near-infrared (NIR) absorption to measure spectral signatures in order to uniquely identify and quantify the analytes of interest (see Simultaneous Measurement of Glucose and in Insect Cell Media Culture by Near Infrared Spectroscopy, Riley et al, Biotechnology and Bioengineering, 55, 1, p. 11, 1997 and Determination of physiological levels of glucose in an aqueous matrix with digitally filtered Fourier transform near-infrared spectra, M. A. Arnold and G. W. Small, Anal. Chem. 62, p. 145, 1990). Attempts to quantify biological materials using NIR (near infra-red) spectroscopy generally fail due to the lack of specificity provided by this type of spectroscopy. The spectral features found in the NIR are generally not particularly sharp (narrow) or distinct, thereby making it difficult to recognize the spectral features, and extract individual analyte concentrations owing to spectral overlap. Specifically, when many analytes are present simultaneously, it is difficult if not impossible even using sophisticated multi-variate computer-based algorithms to obtain clinically accurate results. Additionally, any system noise compromises the spectral data as the broad and rolling spectral features overlap and are smeared out. Even determining the exact spectral peak locations and amplitudes can often be extremely challenging.
More recent spectroscopic investigations have utilized the Raman scattering effect to interrogate the sample, as Raman spectra are generally sharper and more distinct (see Concentration Measurements of Multiple Analytes in Human Sera by Near-Infrared Laser Raman Spectroscopy Jianan Y. Qu, Brian C. Wilson, and David Suria, Applied Optics, 38, 25, p. 5491, 1999 and Rapid, noninvasive concentration measurements of aqueous biological analytes by near-infrared Raman spectroscopy, Andrew J. Berger, Yang Wang, and Michael S. Feld, Applied Optics, Vol. 35, 1, p. 209, 1996). Raman scattering is a known nonlinear optical scattering process which involves inelastic scattering of a photon and an atom or molecule. When light is scattered from an atom or molecule, most photons are elastically scattered such that the scattered photons have the same energy and wavelength (frequency) as the incident photons (i.e., Rayleigh scattering). However, a small fraction of the scattered photons have a frequency different from the frequency of the incident photons. This new frequency can be higher and/or lower, but the scattering to the lower energy level occurs with a far higher probability per unit time. For this inelastic scattering process to proceed, the energy (and momentum) difference is taken up by the atom or molecule in a process whereby an electronic, vibrational, or rotational quantum is excited and/or de-excited.
The inelastic scattering process known as the Raman Effect leads to both lower and higher energy scattered photons, which are referred to as Stokes scattering and anti-Stokes scattering, respectively. The Raman Effect is often modeled as the absorption and subsequent re-emission of a photon via an intermediate vibrational state, having a virtual energy level. If/when this absorption and re-emission of light occurs in Raman scattering there is an energy exchange between the incident photons and the molecules. The energy differences are equal to the difference in the electrical, vibrational or rotational energy-levels of the atom/molecule. In Raman spectroscopy of molecules, the vibrational energy shift is the most commonly observed. If a molecule absorbs energy and a lower energy photon is emitted, it is referred to as Stokes scattering. The resulting photons of lower energy have an energy distribution which generates a Stokes spectrum that is “red-shifted” from the incident beam, or equivalently stated, lower in energy than the pump light. If the molecule loses energy (gives up a quantum unit of energy from an excited electronic, vibrational, or rotational state) by having that energy combine with an incident photon, it is referred to as anti-Stokes scattering. These incident photons are shifted to the higher energy (blue) side of the incident light
These differences in energy between incident and scattered photons are measured by subtracting the energy of the single-frequency excitation laser light source from the energy of the inelastically scattered photons. The intensities of the Raman bands are dependent on the number of molecules occupying the different vibrational states, when the process began. If the sample is in thermal equilibrium, the relative numbers of molecules in states of different energy will be given by the Boltzmann distribution:
            N      1              N      0        =                    g        1                    g        0              ⁢          e              -                              Δ            ⁢                                                  ⁢                          E              v                                kT                    where:                N0: number of atoms the lower vibrational state        N1: number of atoms in higher vibrational state        g0: degeneracy of the lower vibrational state        g1: degeneracy of the higher vibrational state        ΔEv: energy difference between these two vibrational states        k: Boltzmann constant        T: temperature Kelvin        
It can be seen from the Boltzmann relationship shown above that lower energy states will be more highly occupied than the higher energy states. Therefore, the Stokes spectrum will be significantly more intense than the anti-Stokes spectrum generally by a few orders of magnitude. Given that the spontaneous Stokes signal is orders of magnitude lower in power than the incident light, the spontaneous anti-Stokes signal in most free space optical systems will be very close in amplitude to the noise floor of the measurement system. Additionally, the scattered light will be Lambertian and therefore the capture of the scattered light for analysis will be limited by a restatement of the 2nd law of thermodynamics known as conservation of brightness (see Art of Radiometry, James Palmer and Barbara Grant, 2010 SPIE, ISBN 978-0-8194-7245-8). It is for this reason that the anti-Stokes spectrum has typically not been of common or practical use in analyte detection and identification.
The most commonly employed Raman spectroscopy system for analyte identification of biological samples is a Raman Stokes system that uses a laser source having a wavelength of 785 or 830 nm. The excitation wavelength can be anywhere from the ultra violet to the infrared, but the most common wavelengths employed are between 700 nm-900 nm (near infrared or NIR) window because laser sources of this wavelength are readily available and most tissue and biological fluids exhibit minimum absorption in this wavelength region so that auto-fluorescence (which results in a non-uniform baseline shift interference) is reduced. These advantages of the NIR pump region are counter balanced by two limitations:                1. The Raman Stokes scattering cross section, (hence signal intensity) has a 1/λ4 dependence, where λ here represents the wavelength of the excitation source;        2. The sensitivity of the most common multi-channel detectors used in conjunction with NIR optical spectrometers, namely silicon CCDs (charge coupled devices), falls off rapidly for wavelengths exceeding 1 μm.        
Therefore, in the 785 to 830 nm wavelength regime there is a local optimum that can be achieved by a tradeoff between Raman cross-section, auto-fluorescence, detector sensitivity, and filter efficiency. Systems are employed at other wavelengths depending on situation and the wavelength dependence of the Raman cross-section, or if resonance Raman scattering is utilized (see Achievements in resonance Raman spectroscopy: Review of a technique with a distinct analytical chemistry potential, Evtim V. Efremov, Freek Ariese, Cees Gooijer, Analytica Chimica Acta, Volume 606, Issue 2, 14 Jan. 2008, Pages 119-134). This depends on which part of the atomic or molecular (electronic, rotational, vibrational) spectrum which is intended to be probed, as well as the complexity and molecular structure of the analyte species and background matrix being measured.
In practice, art workers implementing optical free space spontaneous Raman spectroscopy systems have generally utilized a carefully designed free space optical system (bulk lenses, reflecting collection systems, or the like) to collect the Raman Stokes scattered light. As previously mentioned, given that the Raman scattered light generally constitutes what is known as a Lambertian source the detection system collection efficiency and coupling to the spectrometer will be limited by the law of the conservation of brightness (see Art of Radiometry, James Palmer and Barbara Grant, 2010 SPIE, ISBN 978-0-8194-7245-8).
The existing body of scientific literature contains descriptions of numerous systems for analyte identification and quantification based on Raman spectroscopy systems. A bio-analysis system was described by Berger et al (Multicomponent Blood Analysis by Near-Infrared Raman Spectroscopy, Andrew J. Berger, Tae-Woong Koo, Irving Itzkan, Gary Horowitz, and Michael S. Feld, Applied Optics, 38, 13, p. 2916, 1999). This particular system was designed to non-invasively measure analytes of medical significance in human blood (e.g.: Glucose, Cholesterol, Triglyceride, Urea, Total Protein, and Albumin). The apparatus utilized a diode laser emitting at 830 nm, a mirror and lens system to collect the Raman Stokes light and deliver it to a spectrometer that employed a silicon CCD array. This system reportedly yielded quantitative results that approached the accuracy required to be used in clinical measurements. Given that the Raman Stokes scattered light levels were low and that they were going through a turbid media (skin and blood), Berger at al. used the method of Partial Least Squares (PLS) and training sets in order to retrieve the concentration data from the spectral data of multiple analytes simultaneously (see Multivariate Calibration, H. Martens and T. Naes, Wiley, New York, 1989 or Mixture analysis of spectral data by multivariate methods, or Windig, W., Chemom. Intell. Lab. Syst. 4, p. 201-213, 1988). Training sets are the response function spectra of the instrument to a known set of analytes with known concentrations. These training sets are required so that the PLS algorithm can create a set of basis vectors that are then used computationally to determine the concentrations of multiple analytes in a given sample's Raman Stokes spectrum. This sample's total spectrum is comprised of the spectra of each of the individual analytes hitting the detector simultaneously and with the relative amplitudes of each individual component of the spectrum determined by their concentration, Raman cross section, the pump light amplitude, and scattering at the respective wavelengths comprising the spectrum. If the conditions change such that the training sets are no longer valid (e.g., as a result of the addition of other spectral components or a change in the spectral baseline), then the results gained using PLS may no longer be correct. This can occur for instance in a biological sample if there is spectral content added due to a bacterial infection, or if the sample is modified in a way that changes the basis vectors of the training set. Multivariate analysis techniques like PLS are also sensitive to changes in both the signal and/or the training sets, and quantitation errors can creep in due to noise or other effects that impact either the signal integrity or the training sets. Also, training sets are in general both time-consuming and can sometimes be difficult to create. Finally, if anything changes the nature of the system by changing the constituent make up or composition, the training sets may no longer be valid, thereby invalidating the basis vectors and signals. For example, such a situation can easily be envisaged in biological systems where an adventitious agent (bacterial of viral) can change the chemical make-up of a system or in chemical systems where dyes or additives can clearly change the absorption profiles of the system.
Another complication for this type of system, and indeed for almost all optical systems including Raman spectroscopy systems used to measure biological samples, is fluorescence. As both the Raman Stokes signal and auto-fluorescence emissions are red-shifted from the pump light, there is always overlap between the two signals. This complicates the detection and identification process as the fluorescence emission often obfuscates the Raman Stokes spectrum. A known technique to try and mitigate the effects of auto-fluorescence is to fit it to a high order polynomial (typically 5th order) and subtract it out of the spectrum (see Automated Method for Subtraction of Fluorescence from Biological Raman Spectra, Lieber and Mahadevan-Jansen, Applied Spectroscopy, 57, 11, p. 1363, 2003). While this technique aids in cleaning up the spectrum, it is not obvious that all of the auto-fluorescence is accurately fitted and that the amplitudes of the spectral features that are revealed are absolutely or even relatively correct. A published example of this fitting and subtraction technique is shown in FIG. 1 (see Quantitative analysis of serum and serum ultrafiltrate by means of Raman spectroscopy, Rohleder et al, Analyist, 129, p. 906, 2004). While the peaks are more visibly revealed, there is no evidence that the amplitudes of all the features have been maintained in true proportion thereby leading to the potential for quantitation errors when used in conjunction with multi-variate analysis.
Clearly, if the Raman Stokes signal can be increased relative to the fluorescence background, many of these issues can be mitigated. However, the Raman Stokes signal and the fluorescence signal both vary linearly with the intensity of the pump light so it is difficult to preferentially generate a Stokes signal with higher relative amplitude to the fluorescence. The act of subtracting off a generic function as described above can often be the mathematical operation of subtracting signals of similar magnitude and thereby offers little improvement in the signal to noise ratio and adds uncertainty to the resulting spectrum. However, in samples that are first passed through a filtration/optical scattering particle reduction system (e.g.: ultrafiltration, centrifugation etc.) the increased Raman Stokes amplitude can result in higher signal to noise ratios as the filtering reduces the auto-fluorescence and scattering losses (see Quantitative analysis of serum and serum ultrafiltrate by means of Raman spectroscopy, Rohleder et al., 129, p. 906, Analyst, 2004). Despite these facts, many attempts have been made to increase the Raman signal including using higher power pump lasers, and increasing the density of the target analyte. Some approaches have generally utilized clever optical systems to preferentially capture more of the Raman Stokes scattered signal light.
Hollow waveguide technology is one method for increasing the Raman Stokes signal that holds promise. The use of hollow core Teflon AF® tubing as an optical waveguide was met with great interest when first demonstrated by Altkorn (see Low-loss liquid-core optical fiber for low refractive index liquids: fabrication, characterization, and application in Raman spectroscopy, Alkorn et al, Applied Optics, 36, 34, p. 8992, 1997). Teflon AF is one of the few materials with a refractive index lower (n˜1.29) than many aqueous solutions (nwater˜1.33) a property that allows it to serve as an optical waveguide. This fact allows many aqueous solutions to be analyzed using Raman spectroscopy by introducing the solution into the hollow core Teflon AF waveguide. The increased confinement in the core and increased interaction length both act to enhance the Raman signal. Specifically, the Teflon tubing acts as a waveguide as the pump rays introduced into the core and the Raman Stokes signal generated in the core undergo total internal reflection at the liquid/Teflon boundary and are thereby confined primarily within the core. This results in an increased interaction path and a higher level of intensity over this path than a comparable free space system. Increases in the sensitivity of the system by a factor of 500 have been reported, though the general enhancement factor that can be achieved is correlated to the exact experimental geometry implemented (see Intensity Considerations in Liquid Core Optical Fiber Raman Spectroscopy, Altkorn et al, Applied Spectroscopy, 55, 4, p. 373, 2001 and Raman Sensitivity Enhancement for Aqueous Protein Samples Using a Liquid-Core Optical-Fiber Cell, M. J. Pelletier and Altkorn, Anal. Chem., 73 (6), pp 1393-1397, 2001)
Unfortunately, small diameter (sub 100 micron inner-diameter) Teflon capillary tubing is not readily available and it is therefore difficult to make a single mode waveguide in the near-infrared spectral region. This is because the number of propagating modes in this type of waveguide is dependent on the product of the ratio of the core diameter to the wavelength and the square root of the difference between the squares of the core refractive index and cladding refractive index. Often with the case of Teflon waveguides, the waveguide diameter approaches a level where it is two orders of magnitude larger than the wavelength of the light propagating so that the mode picture is replaced by a ray optics and numerical aperture description of the light propagation within the tube (see Intensity Considerations in Liquid Core Optical Fiber Raman Spectroscopy, Altkorn et al, Applied Spectroscopy, 55, 4, p. 373, 2001). Moreover Teflon materials can be difficult to work with as the low surface tension does not allow them to readily bond with other materials. The larger diameter tubing used in the literature has resulted in waveguides that are not single mode for either NIR pump wavelengths or the Raman signal. The fact that the waveguide is multi-spatial mode allows the pump and the Raman Stokes signal to have different spatial profiles and effective velocities in the waveguide thereby limiting the integrated spatial overlap and overall conversion efficiency from the pump to the Raman Stokes signal. Additionally, as Teflon is difficult to bond to, the coupling of the light and fluid into and out of the fiber is usually accomplished using mechanical fixturing (as opposed to an integrated set of bonded components) which can be cumbersome to implement. Finally, there are fundamental limitations to the density of materials that can be analyzed with Teflon tubing, as the refractive index of the material approaches that of the Teflon cladding.
Another impediment to accurate quantification in the identification of biological or other types of samples with optical analyzers based on Raman scattering is the linear optical loss caused by particles in the sample and/or absorption of the pump or the Raman scattered light by the sample. For instance, in whole blood, there is both absorption and scattering in the NIR. The blood cells can create large scattering losses and the Raman Stokes levels can be mediated by direct scattering of the Raman Stokes signal and/or by pump light scattering as well as by pump light and/or signal absorption. The effect of absorption or scattering loss can be accounted for if the magnitude of the loss coefficient is known, but this is often very difficult to determine a priori. Although one can devise an instrument to measure the loss coefficients in-situ the overall instrument set-up becomes quite complicated and therefore impractical to employ commercially. An example of a laboratory system where this has been implemented is shown in FIG. 3 (see Chemical concentration measurement in blood serum and urine samples using liquid-core optical fiber Raman Spectroscopy, Qi and Berger, Applied Optics, 46, 10, p. 1726, 2007). Here a Raman Stokes analyzer using a Teflon AF waveguide and simultaneously employed a white light spectrometer to account for the scattering and absorption in the sample along the waveguide path. The authors here also reported viable results using this system, but required integration times in excess of 10 seconds despite the waveguide enhancement and the compensation for optical losses.
Some of the aforementioned issues with Teflon based hollow core fibers have been somewhat overcome with the advent of hollow core photonic band-gap (HCPBG) fibers (see Photonic Crystal Fiber, Philip Russell, Science 17, 299, 5605, p. 358, 2003, U.S. Pat. No. 6,829,421, Hollow Core Photonic Bandgap Optical Fiber, and Published US Patent Application 2006/0062533, Photonic Crystal Fiber, Method of Manufacturing the Crystal Fiber and Method of Connecting the Fiber). The nature of the photonic band-gap allows most gases or liquids to be confined in the core of the fiber, while guiding of the excitation (pump) and scattered (signal) light is supported. Additionally, it allows for the fiber to be constructed so that the sample of interest is introduced into the core but the fiber remains single mode or close to single mode for both the pump light and the emitted spectra in the near infrared spectral region. Several groups have implemented Raman Stokes analyzers using HCPBG fiber and others have explored various systems and concepts using this type of fiber (see Published US Patent Application 2010/0014077, U.S. Pat. No. 7,595,882, Stimulated Raman scattering in an ethanol core microstructured optical fiber, Yiou et al, Optics Express, 13, 12, p. 4786, 2005, and Determination of Ethanol Concentration by Raman Spectroscopy in Liquid-Core Microstructured Optical Fiber, Meneghini et al, IEEE Sensors Journal, 8, 7, p. 1250, 2008). U.S. Pat. No. 7,595,882 describes a system for identifying homo-nuclear molecules confined to the core of HCPBG fibers. This patent describes the general advantages provided by confining the radiation and the sample to a hollow core. Published US Patent Application 2010/0014077 describes a method for identifying biological samples using HCPBG as a Raman biosensor. In addition to a general description of how to use the Raman Stokes signal to create a bio-analyzer, this patent application discusses how to pick the fiber core diameter or the wavelength of the pump light for a given fiber design and a given analyte mixture's index of refraction based on a published reference. If a commercially available HCPBG fiber comes in specific discrete core diameters, the single-mode cut-off wavelength is determined for a given index of refraction media in the hollow-core based on the photonic band gap cladding. The above-cited patent application focuses solely on Raman Stokes signals as indicated by the spectra shown with a positive wavelength shift and makes no mention of an anti-Stokes spectrum. We have found that stimulated Raman anti-Stokes based spectroscopy is also possible with HCPBG fiber. The first stimulated Raman anti-Stokes signal will occur when there is a large enough population build-up in the first Raman Stokes and Raman anti-Stokes levels to allow for this process.
Recent work was performed by Meneghini et al. (see Determination of Ethanol Concentration by Raman Spectroscopy in Liquid-Core Microstructured Optical Fiber, Meneghini et al, IEEE Sensors Journal, 8, 7, p. 1250, 2008). Here a HCPBG fiber was used to determine ethanol and sucrose content in a set of samples. In the described system, the HCPBG fiber was spliced to multi-mode fiber on both the input and output ends so that pump light could be coupled in and Raman Stokes light could be coupled out and sent to a spectrometer. The input and output fibers were attached to the HCPBG fiber using a fusion splicer. The act of fusion splicing also collapsed the cladding holes around the launch (light input) area. Additionally, laser holes were drilled into the side of the fiber allowing the core to be filled with the ethanol and sucrose containing solutions that comprised the samples under test. The system reportedly gave clean and well resolved Raman Stokes spectra, thereby allowing for reasonably accurate quantification of the analyte concentrations using univariate techniques. Spectral graphs were obtained with 1 mW and 1 meter of HCPBG fiber as there was no fluorescence to complicate the spectrum and little scattering. Their testing, however, did not involve biological samples and/or any samples having large fluorescence and/or scattering/absorptive backgrounds. Additionally, it was clear from their work that very low pump laser powers (<10 mW) and very short fibers (<2 M) were required to get their results.
It is important to note that while all of the previously referenced prior art refers to “Raman scattering”, the term was invariably used solely in the context of the Raman Stokes spectrum The only reported work known to the present inventors which addresses identifying analytes using the Raman anti-Stokes spectrum is described in Biological Applications of Anti-Stokes Raman Spectroscopy: Quantitative Analysis of Glucose in Plasma and Serum by Highly Sensitive Multichannel Raman Spectrometer, Dou et al, Applied Spectroscopy, 50, 10, p. 1301, 1996. It was reported by Dou et al, that by employing the complex free space optical collection system they utilized, the anti-Stokes spectrum could be used to predict the concentration of glucose in blood serum if their collection system was implemented. Their optical collection system (as shown in FIG. 3) consisted of a quartz flow cell surrounded by a gold coated integrating ellipse (it collects or integrates the signal) with two holes in it, one small hole 310 that allowed the Argon Ion pump laser light to pass into the cell, and a small conically shaped hole 320 through which the anti-Stokes spectrum escaped and was subsequently collimated and sent to a holographic spectrometer and CCD array. The elliptical chamber 315 was designed to optimize the Raman anti-Stokes signal emanating from the quartz flow cell 300. Given the very short interaction length of the pump and the analyte and the fact that the spontaneous Raman anti-Stokes signal is emitted into all space, efficient generation and collection of the signal is very challenging. Despite their carefully designed and implemented free space collection system, they still required 100 mW of pump power at 5145 nm and also needed to integrate on their CCD for 3 seconds to obtain reasonably clear spectra. Also, they simultaneously collected both the Stokes and anti-Stokes spectrum in the presence of heavy fluorescence of the plasma (as shown in FIG. 4). Using the Raman anti-Stokes band at 1130 cm−1 they were able to create a plot of glucose concentration in blood plasma that matched the intensity on the CCD with a correlation coefficient of 0.993; as is shown in FIG. 5. They apparently did not require multivariate analysis/PLS and/or training sets in order to determine the concentration of the analyte that was varied. However, it should be noted that if multiple target analytes are present, there is significant potential for “interferences” (i.e. overlaps between the Raman anti-Stokes spectra of the various target analytes). Depending on the analytes that comprise each sample mixture, investigation into the actual or potential overlaps must be carried out in order to map out these issues and mathematically determine the spectra and peaks as a function of each target analyte concentration or as a function of the presence of multiple analytes. Additionally, the Raman anti-Stokes peaks used to determine the concentration of the analyte need to be directly correlated with the concentration. In general one would require multiple peaks to generate the correlation or correlation function to the concentration.